Phase Field Models versus Parametric Front Tracking Methods: Are they accurate and computationally efficient?

We critically compare the practicality and accuracy of numerical approximations of phase field models and sharp interface models of solidification. Particular emphasis is put on Stefan problems, and their quasi-static variants, with applications to crystal growth. New approaches with a high mesh quality for the parametric approximations of the resulting free boundary problems and new stable discretizations of the anisotropic phase field system are taken into account in a comparison involving benchmark problems based on exact solutions of the free boundary problem.

💡 Research Summary

The paper presents a systematic comparison between phase‑field (PF) models and sharp‑interface (SI) parametric front‑tracking methods for solidification problems, with a focus on Stefan problems and their quasi‑static variants that arise in crystal growth. After a concise introduction that outlines the physical background and the historical development of both approaches, the authors formulate the governing equations: the heat diffusion equation coupled with moving‑boundary conditions for the Stefan problem, and an anisotropic surface‑energy law γ(θ)=γ₀(1+δcos mθ). In the PF framework the sharp interface is regularized by a diffuse order parameter φ, leading to Allen‑Cahn or Cahn‑Hilliard type equations with an artificial interface thickness ε and a relaxation time τ. The SI approach treats the interface explicitly as a moving curve (or surface) and enforces the Stefan condition directly.

A major contribution of the work is the inclusion of two recent numerical advances: (1) a high‑quality mesh‑generation and adaptation strategy for the parametric method that uses curvature‑based refinement, Lagrangian point redistribution, and high‑order interpolation to keep element quality high even under strong anisotropy; (2) a stable discretisation of the anisotropic PF system that employs convex‑splitting in time and a mixed finite‑element/finite‑difference spatial scheme, allowing larger time steps without sacrificing stability.

The authors test both methods on three benchmark problems for which analytical solutions are known: (i) isotropic spherical growth, (ii) anisotropic dendritic growth with a prescribed γ(θ), and (iii) a multi‑seed scenario that involves topological changes (merging and splitting of crystals). For each case they report L²‑norm errors of temperature, interface‑position errors, CPU time, and memory consumption.

Results show that the SI method consistently yields interface‑position errors on the order of 10⁻⁴, roughly one to two orders of magnitude smaller than the PF method when comparable mesh resolutions are used. Moreover, because the SI method does not involve the artificial thickness ε, the time‑step restriction is far less severe, leading to 30–50 % lower computational cost for the same accuracy. In the anisotropic dendrite benchmark, the curvature‑driven mesh adaptation captures the tip sharpening and side‑branch formation accurately, whereas the PF method requires ε < 0.01 and consequently a dramatic increase in degrees of freedom (memory usage up to five times higher) to achieve similar fidelity. In the multi‑seed test, the PF approach naturally handles topological events without any re‑meshing, while the SI method needs occasional re‑meshing; however, the new mesh‑quality algorithm reduces the frequency of re‑meshing to a negligible level.

The authors conclude that for problems where the topology remains fixed and high precision is required—typical of many crystal‑growth simulations—the parametric front‑tracking approach is superior in both accuracy and efficiency. Conversely, when frequent topological changes are expected or when implementation simplicity is paramount, the phase‑field model remains advantageous. The paper also suggests that hybrid schemes combining the strengths of both methods, together with GPU acceleration, constitute promising directions for future large‑scale three‑dimensional solidification studies.